Originally posted by pingflood So how much impact does the more compact/dense circuitry of an APS-C or 4/3 sensor have on noise compared to FF (assuming same pixel count)?
This is another famous debate
I prefer to discuss this topic independent of sensor size because the size of a pixel and the size of the sensor are not linked to each other. It is just to simplify the discussion as in this thread that one keeps constant the pixel count constant.
If you take the D300, you see two FF derivatives: D3 (same pixel count) and D3X (same pixel pitch).
So, the question really is how much impact does the pixel pitch have on noise, independent of sensor size?
Here, Haakan's nice reinterpretation of DxO means that the impact in none.
In general, to compare noise, you must first normalize it to a given print size (factor out the different number of pixels resulting from the different pixel pitch). E.g., in order to normalize to 10 MPixels, you either resize all samples to this size, or multiply the ISO of measurements by SQRT(pixels/10mp) in ISO (which is a simple shift in ISO on a logarithmic ISO scale). I.e., You compare a 14.6mp/ISO800 pixel measurement with 10mp/ISO1000.
This normalization removes the first order effect of pixel pitch, i.e., the scale dependence of noise.
The second order effects are more subtle. On the one hand, you have more circuitry overhead per pixel (a larger shaded area), smaller circuitry (more electronic noise), etc. On the other hand, you have more pixels to average out the noise coming from, e.g., dark current or gain amplifiers. And microlenses reduce the effect of shading.
The second order effects are such that there exists a sweet spot (optimum pixel size) resulting in minimum normalized noise, which depends on the silicon process, chip and microlens design, and sensor temperature. Experiments with RAW files from P&S cameras suggest that this sweet spot is at somewhat smaller pixel sizes than in current use in dSLRs.
Normally, large pixel pitch FF sensors are expensive enough to have carefully designed gain amplifiers and ADC circuitry not adding to normalized noise.
If you compare D3 and D3X in DxO mark's SNR plot (the "Print" does the scale normalization and must be looked at), then you see almost identical results (less than 1dB difference).
So, to summarize: The pixel pitch can be ignored in this discussion. Fortunately.
--
Personally, I like the resolution advantage coming from high pixel pitch FF (like D3X). But it is expensive from Nikon (somewhat more affordable with 5D2 and esp. A850). Where high pixel pitch FF currently means 6µm vs. 5µm in K-7 and 4.3µm in 7D. And a high pixel pitch FF makes video expensive to implement if you don't want a noise penalty in video.